The Impact of Multiple Parallel Phrase Suggestions on Email Input and Composition Behaviour of Native and Non-Native English Writers

Buschek, Daniel; Zürn, Martin; Eiband, Malin

doi:10.1145/3411764.3445372

Cited by 62 publications

(46 citation statements)

References 69 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Who will own the authorship of the model-generated content and how could the shifting responsibilities and ownership of the consent be misused [Weiner 2018] (see §5.5: economics for a more in-depth discussion)? What are the long-term implications that foundation models will have on our work, language and culture [Hancock et al 2020;Buschek et al 2021]? Of particular relevance to this last question is the fact that foundation models are trained on observed data and do not necessarily inform us about causality.…”

Section: Impact On End-user Interaction With Ai-infused Applicationsmentioning

confidence: 99%

On the Opportunities and Risks of Foundation Models

Bommasani¹,

Hudson²,

Adeli³

et al. 2021

Preprint

649

652

View full text Add to dashboard Cite

AI is undergoing a paradigm shift with the rise of models (e.g., BERT, DALL-E, GPT-3) that are trained on broad data at scale and are adaptable to a wide range of downstream tasks. We call these models foundation models to underscore their critically central yet incomplete character. This report provides a thorough account of the opportunities and risks of foundation models, ranging from their capabilities (e.g., language, vision, robotics, reasoning, human interaction) and technical principles (e.g., model architectures, training procedures, data, systems, security, evaluation, theory) to their applications (e.g., law, healthcare, education) and societal impact (e.g., inequity, misuse, economic and environmental impact, legal and ethical considerations). Though foundation models are based on standard deep learning and transfer learning, their scale results in new emergent capabilities, and their effectiveness across so many tasks incentivizes homogenization. Homogenization provides powerful leverage but demands caution, as the defects of the foundation model are inherited by all the adapted models downstream. Despite the impending widespread deployment of foundation models, we currently lack a clear understanding of how they work, when they fail, and what they are even capable of due to their emergent properties. To tackle these questions, we believe much of the critical research on foundation models will require deep interdisciplinary collaboration commensurate with their fundamentally sociotechnical nature.

show abstract

Section: Impact On End-user Interaction With Ai-infused Applicationsmentioning

confidence: 99%

On the Opportunities and Risks of Foundation Models

Bommasani¹,

Hudson²,

Adeli³

et al. 2021

Preprint

649

652

View full text Add to dashboard Cite

show abstract

“…Additionally, in a study by Robertson et al [92], the authors found that instead of accepting an AI-generated text suggestion by clicking on it, one participant chose to write out the same words themselves, expressing that they desired more agency when writing messages. Buschek et al [18] also demonstrated that the more AI-generated text suggestions a person accepted in their emails, the lower they perceived their authorship over the message to be. Both of these insights emphasize the action initiation dimension of agency; participants were seemingly interested in feeling like they were the ones who generated the text in their messages, as opposed to the AI system.…”

Section: Human and Machine Agencymentioning

confidence: 99%

Examining Agency, Expertise, and Roles of AI Systems in AI-Mediated Communication

Mieczkowski¹,

Hancock²

2022

Preprint

View full text Add to dashboard Cite

AI-Mediated Communication (AI-MC) occurs when an AI system operates on behalf of an individual in communication between people. Building on literature in psychology and human-machine communication, the following three studies answered two overarching questions: 1) how do people perceive their own agency when an AI system operates on their behalf in interpersonal communication? and 2) how do people perceive the role(s) of the system under these circumstances? Through a think-aloud study, as well as an online experiment, these studies revealed that people strive to maintain their agency in AI-MC, and devise various strategies to do so. Additionally, even though AI systems can play many roles in a communication task, people view them as social actors instead of as extensions of the self, which impacts how they attribute agency and behave in human-AI interactions. Lastly, people show increased reliance on the AI system when they lack expertise, which underlies decreased feelings of agency as well as changes in collaboration dynamics. Taken together, these studies apply research on agency and human-AI interaction to the nascent study of AI-MC. The results provide insights into the effects of AI-involvement and topical expertise on people’s perceptions and behaviors, as well as how they think about the roles of AI systems.

show abstract

“…For instance, Todi et al (2021) showed that approaches based on reinforcement learning can be used to automatically adapt related user interfaces. For interactive NLP, Buschek et al (2021) investigated how different numbers of phrase suggestions from a neural language model impact user behavior while writing, collecting a dataset of 156 people's interactions. In the future, data such as this might be used, for example, to train a model that replicates users' selection strategies for text suggestions from an NLP system.…”

Section: Employing User Models As Proxies For Interactive Evaluationsmentioning

confidence: 99%

Methods for the Design and Evaluation of HCI+NLP Systems

Heuer,

Buschek

2021

Preprint

Self Cite

View full text Add to dashboard Cite

HCI and NLP traditionally focus on different evaluation methods. While HCI involves a small number of people directly and deeply, NLP traditionally relies on standardized benchmark evaluations that involve a larger number of people indirectly. We present five methodological proposals at the intersection of HCI and NLP and situate them in the context of ML-based NLP models. Our goal is to foster interdisciplinary collaboration and progress in both fields by emphasizing what the fields can learn from each other.

show abstract

The Impact of Multiple Parallel Phrase Suggestions on Email Input and Composition Behaviour of Native and Non-Native English Writers

Cited by 62 publications

References 69 publications

On the Opportunities and Risks of Foundation Models

On the Opportunities and Risks of Foundation Models

Examining Agency, Expertise, and Roles of AI Systems in AI-Mediated Communication

Methods for the Design and Evaluation of HCI+NLP Systems

Contact Info

Product

Resources

About